Evolution as a guide for autonomous vehicle path planning and coordination

Influenced by the problem-solving capability observed in nature, we utilize the mechanism of simulated evolution as the basis for path-planning algorithms for uninhabited vehicles. An attractive feature of evolution-based search is the general nature of performance criteria that can be defined, incorporating both continuous and discrete measures of candidate solutions. These criteria can often be defined as straightforward mathematical realizations of "fuzzy" objectives, consistent with the use of natural language syntax and dialogue. This paper investigates the influence of the population representation on the efficacy of solution of a series of static path planning problems of increasing complexity. Furthermore, we demonstrate the potential for solving mission planning problems at higher levels of abstraction by extending the evolution-based formulation to handle coordination of multiple vehicles. We apply this framework to a multiple traveling salesperson problem in which the vehicles work together to accomplish goals with respect to an overall team performance metric.